This proposal entitled "Correct Alignments and Improved Refinements for High-Accuracy Structure Modeling" follows closely the two goals defined by the NIH RFA, namely, (1) getting crystal structure quality models for close homologs (more than 30% sequence identity) and (2) building high accuracy models for remote homologs (as low as 10% sequence identity). We believe that reaching these goals will require the development of (a) new approaches to integrate all biological and structural information available on a protein sequence family to improve the alignment of the target protein sequence to a known structural template, (2) new methods for sampling the conformational space accessible to a protein structure, and (3) new methods that provide accurate refinements of near native protein structural models. We will achieve these goals through the following specific aims. (1) Generate accurate alignments between the target protein sequence and a structural template by combining a wide range of different sources of information. Essential to this aim is the integration of these data into a unified framework. We will develop the concept of residue position annotation (RPA) for homology modeling, in which different positions in the sequence have different impact in the modeling procedure depending on their properties, derived from multiple sequence alignments. We will use the framework of mean field minimization to design a new alignment package that incorporate different types of constraints, even non-additive. (2) Build high accuracy models for the target protein. Homology modeling based on a given sequence alignment between a target sequence and the sequence of a template protein whose structure is known usually involve two steps: loop building, to fix regions of insertion and deletion in the alignment, and side-chain modeling. We will elaborate from our extensive experience in developing solutions to both problems to propose new approaches that circumvent these difficulties. (3) Improve initial models to generate crystal structure quality models using structure refinement. We will follow three directions (a) minimization and molecular dynamics with improved force fields, including quantum mechanical terms, (b) minimization and molecular dynamics with improved implicit solvent models and (c) energy minimization with cooperative many-body energy terms. (4) Organize all computer programs developed within this proposal into a user-centric package, including visual computing tools, to make it accessible to the biological community at large.